Purpose
To develop a technique based on image registration to correct transverse motion artifacts in OCT images after scan acquisition.
Methods
The OCT system mainly consists of a commercially available femto-second Ti:Sapphire laser unit (FEMTOLASERS Produktions GmbH, AUSTRIA, Spectral Range : 730 nm - 940 nm) as a light source, a original spectrometer, and a computer for data acquisition and image processing. The OCT system can create cross-sectional images (B-scans) of retinal structure with an axial resolution of 2.0 μm and an imaging speed of 37,000 single axial scans (A-scans) per second. Multiple B-scans through the central fovea were carried out to obtain multiple OCT images. The first stage involved selecting a base image that was suitable for the target image averaging process from all the available images. Moreover, multiple images which have high correlation with the base image were selected. A time averaged images was created within the next stage using a line-divided pattern matching technique from the selected images. The technique comprised the search for a line on the selected image that corresponded to the reference line on the base image. Two lines on the base and selected images, having a similar texture in their respective lines of interest, were regarded as the corresponding lines. The similarity was measured by the cross correlation coefficient.
Results
Nineteen OCT images were taken from 9 normal subjects. The accuracy of the image registration process can be evaluated by calculating correlation coefficient between the base image and selected images. The correlation coefficient was 0.71±0.04, which was higher than 0.54±0.054 when simple whole-area registration was applied (P<0.0001, paired t-test).
Conclusions
This technique offers a simple yet effective procedure to significantly improve visualization of OCT image for clinical diagnosis.
Keywords: 549 image processing •
552 imaging methods (CT, FA, ICG, MRI, OCT, RTA, SLO, ultrasound)